AutoAdy's MCP server is not Claude-only. It speaks the open Model Context Protocol over streamable HTTP, and it is one endpoint with one tool catalogue, whatever is asking.
Grok web and Grok Build document custom remote MCP connections. Grok Bot has a separate plugin flow. Follow the current Settings → Agents guide for your exact client, account sign-in and a read-only connection check.
Grok Bot, Grok Build, personal Muse and Muse Code have different connection flows. The agent guide includes exact config, account sign-in, a read-only connection check and official sources.
Open Grok setup guideGrok web, Grok Build and Grok Bot have separate setup paths and workspace policies. The agent guide links xAI's official documentation and labels the Bot route as conditional. These recipes are documentation-matched; live client sign-in has not been tested here.
“Run a full audit of my ad account and tell me what needs attention.”
calls run_full_audit
“Why did my cost per lead double last week?”
calls diagnose_drop
“Show my top performing ads from the last 30 days.”
calls get_top_performing_ads
“Which ads should I pause? Show me the list before anything happens.”
calls list_kill_recommendations
“Scale my best campaign by 20% without resetting learning.”
calls scale_budget_relative
xAI documents remote MCP for Grok web, Grok Build and API agents. Grok Bot uses a separate custom Remote HTTPS plugin flow, subject to app and team permissions. See Settings → Agents for exact steps, official sources and a read-only test.
No. There is one AutoAdy MCP server at https://www.autoady.io/api/mcp and every assistant connects to it. Same 136 tools, same plan gating, same write cap, same journal. The only difference between these pages is how each client is configured.
Read your Meta ad account in depth (audits, per-campaign and per-ad performance counted against each campaign's own objective, drop diagnosis, creative and fatigue analysis, the open findings queue) and, on a paid plan, change it: budgets, pausing, duplication, targeting and launching. The full generated list is on the Meta Ads connector page.
The nine tools that stop live delivery, seven for Meta and two for Google Ads, are annotated destructive, so a well-behaved client asks first. The Google ones refuse with GOOGLE_WRITES_DISABLED until Google's OAuth verification completes. Underneath that, they need a paid plan, they are capped at 200 account changes per hour, and every change is journalled with its evidence. A pause is reversible from AutoAdy or Ads Manager; archiving a campaign, ad set or ad is not, so it asks for a typed confirmation. The kill list has a read-only twin so the whole set can be reviewed before it runs.
On AutoAdy's servers, and nowhere else. You connect Meta once inside AutoAdy. The assistant calls a tool, and the per-account token is resolved server-side for that call. The assistant never sees it.